Ye, Chenglong

Project details:

With the rising high dimensional data from medical and economics area, traditional statistical analysis methods are incompetent. For high dimensional data, we assume only a small portion of the explanatory variables are related to the response variable. A common strategy for analyzing high dimensional data is variable screening, in the hope of reducing dimensionality to classical low dimensional case.  Unlike variable selection which generally require model specification, variable screening can be achieved by dependence measures and is model-free. There is a growing development of such powerful dependence measures in high dimensional variable screening. However, one common issue in application is that these methods would miss the variables that not marginally but jointly related to the response variable condition on other variables. Inspired by this observation, we propose a new independence measure that measures the additional contribution of a predictor to the mean of response variable, which conditions on other variables. In addition, our proposed measure has the sure screening property, which enable our method to capture all the important variables after screening with large sample size. 

Computational Method:

Submit R batch under linux

Software:

R

Cpu Cores need: 200  

Students:

Collaborators:

Grants:

Publications: